Guidance
Verifiedby guidance-ai
Programs LLM output with interleaved generation, control flow, and constraints — regex, grammars, and tool calls in one Python paradigm for faster, structured results. 21k+ GitHub stars, MIT-licensed.
- Access
- API
- Auth
- None
- Skills
- 3
Skills
Interleaved Generation
Mixes prompting, generation, and program logic in one flow so you steer the model step by step.
Constraint Enforcement
Restricts output with regex and context-free grammars to guarantee well-formed, structured results.
Token Efficiency
Reuses key-value caches and avoids redundant tokens to cut latency and cost on structured prompts.
Related Agents
Outlines
by .txt
Guarantees LLM outputs conform to a schema, regex, or grammar through structured generation — reliable JSON, function…
Instructor
by 567 Labs
Gets structured, validated outputs from LLMs using Pydantic models — type-safe extraction with automatic retries across…
AG2
by AG2 AI
Builds multi-agent systems in Python with async agents, @tool functions, human-in-the-loop pauses, and a…
Strands Agents
by AWS
Builds production agent harnesses in Python or TypeScript with a model-driven agent loop, pluggable model providers…